1 // SPDX-License-Identifier: GPL-2.0+
3 * the_nilfs.c - the_nilfs shared structure.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
11 #include <linux/buffer_head.h>
12 #include <linux/slab.h>
13 #include <linux/blkdev.h>
14 #include <linux/backing-dev.h>
15 #include <linux/random.h>
16 #include <linux/crc32.h>
26 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
28 void nilfs_set_last_segment(struct the_nilfs *nilfs,
29 sector_t start_blocknr, u64 seq, __u64 cno)
31 spin_lock(&nilfs->ns_last_segment_lock);
32 nilfs->ns_last_pseg = start_blocknr;
33 nilfs->ns_last_seq = seq;
34 nilfs->ns_last_cno = cno;
36 if (!nilfs_sb_dirty(nilfs)) {
37 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
40 set_nilfs_sb_dirty(nilfs);
42 nilfs->ns_prev_seq = nilfs->ns_last_seq;
45 spin_unlock(&nilfs->ns_last_segment_lock);
49 * alloc_nilfs - allocate a nilfs object
50 * @sb: super block instance
52 * Return Value: On success, pointer to the_nilfs is returned.
53 * On error, NULL is returned.
55 struct the_nilfs *alloc_nilfs(struct super_block *sb)
57 struct the_nilfs *nilfs;
59 nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
64 nilfs->ns_bdev = sb->s_bdev;
65 atomic_set(&nilfs->ns_ndirtyblks, 0);
66 init_rwsem(&nilfs->ns_sem);
67 mutex_init(&nilfs->ns_snapshot_mount_mutex);
68 INIT_LIST_HEAD(&nilfs->ns_dirty_files);
69 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
70 spin_lock_init(&nilfs->ns_inode_lock);
71 spin_lock_init(&nilfs->ns_next_gen_lock);
72 spin_lock_init(&nilfs->ns_last_segment_lock);
73 nilfs->ns_cptree = RB_ROOT;
74 spin_lock_init(&nilfs->ns_cptree_lock);
75 init_rwsem(&nilfs->ns_segctor_sem);
76 nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
82 * destroy_nilfs - destroy nilfs object
83 * @nilfs: nilfs object to be released
85 void destroy_nilfs(struct the_nilfs *nilfs)
88 if (nilfs_init(nilfs)) {
89 nilfs_sysfs_delete_device_group(nilfs);
90 brelse(nilfs->ns_sbh[0]);
91 brelse(nilfs->ns_sbh[1]);
96 static int nilfs_load_super_root(struct the_nilfs *nilfs,
97 struct super_block *sb, sector_t sr_block)
99 struct buffer_head *bh_sr;
100 struct nilfs_super_root *raw_sr;
101 struct nilfs_super_block **sbp = nilfs->ns_sbp;
102 struct nilfs_inode *rawi;
103 unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
104 unsigned int inode_size;
107 err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
111 down_read(&nilfs->ns_sem);
112 dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
113 checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
114 segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
115 up_read(&nilfs->ns_sem);
117 inode_size = nilfs->ns_inode_size;
119 rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
120 err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
124 rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
125 err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
129 rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
130 err = nilfs_sufile_read(sb, segment_usage_size, rawi,
135 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
136 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
143 iput(nilfs->ns_cpfile);
150 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
152 memset(ri, 0, sizeof(*ri));
153 INIT_LIST_HEAD(&ri->ri_used_segments);
156 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
158 nilfs_dispose_segment_list(&ri->ri_used_segments);
162 * nilfs_store_log_cursor - load log cursor from a super block
163 * @nilfs: nilfs object
164 * @sbp: buffer storing super block to be read
166 * nilfs_store_log_cursor() reads the last position of the log
167 * containing a super root from a given super block, and initializes
168 * relevant information on the nilfs object preparatory for log
169 * scanning and recovery.
171 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
172 struct nilfs_super_block *sbp)
176 nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
177 nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
178 nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
180 nilfs->ns_prev_seq = nilfs->ns_last_seq;
181 nilfs->ns_seg_seq = nilfs->ns_last_seq;
183 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
184 nilfs->ns_cno = nilfs->ns_last_cno + 1;
185 if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
186 nilfs_err(nilfs->ns_sb,
187 "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
188 (unsigned long long)nilfs->ns_segnum,
189 nilfs->ns_nsegments);
196 * load_nilfs - load and recover the nilfs
197 * @nilfs: the_nilfs structure to be released
198 * @sb: super block isntance used to recover past segment
200 * load_nilfs() searches and load the latest super root,
201 * attaches the last segment, and does recovery if needed.
202 * The caller must call this exclusively for simultaneous mounts.
204 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
206 struct nilfs_recovery_info ri;
207 unsigned int s_flags = sb->s_flags;
208 int really_read_only = bdev_read_only(nilfs->ns_bdev);
209 int valid_fs = nilfs_valid_fs(nilfs);
213 nilfs_warn(sb, "mounting unchecked fs");
214 if (s_flags & SB_RDONLY) {
216 "recovery required for readonly filesystem");
218 "write access will be enabled during recovery");
222 nilfs_init_recovery_info(&ri);
224 err = nilfs_search_super_root(nilfs, &ri);
226 struct nilfs_super_block **sbp = nilfs->ns_sbp;
232 if (!nilfs_valid_sb(sbp[1])) {
234 "unable to fall back to spare super block");
237 nilfs_info(sb, "trying rollback from an earlier position");
240 * restore super block with its spare and reconfigure
241 * relevant states of the nilfs object.
243 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
244 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
245 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
247 /* verify consistency between two super blocks */
248 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
249 if (blocksize != nilfs->ns_blocksize) {
251 "blocksize differs between two super blocks (%d != %d)",
252 blocksize, nilfs->ns_blocksize);
256 err = nilfs_store_log_cursor(nilfs, sbp[0]);
260 /* drop clean flag to allow roll-forward and recovery */
261 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
264 err = nilfs_search_super_root(nilfs, &ri);
269 err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
271 nilfs_err(sb, "error %d while loading super root", err);
278 if (s_flags & SB_RDONLY) {
281 if (nilfs_test_opt(nilfs, NORECOVERY)) {
283 "norecovery option specified, skipping roll-forward recovery");
286 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
287 ~NILFS_FEATURE_COMPAT_RO_SUPP;
290 "couldn't proceed with recovery because of unsupported optional features (%llx)",
291 (unsigned long long)features);
295 if (really_read_only) {
297 "write access unavailable, cannot proceed");
301 sb->s_flags &= ~SB_RDONLY;
302 } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
304 "recovery cancelled because norecovery option was specified for a read/write mount");
309 err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
313 down_write(&nilfs->ns_sem);
314 nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
315 err = nilfs_cleanup_super(sb);
316 up_write(&nilfs->ns_sem);
320 "error %d updating super block. recovery unfinished.",
324 nilfs_info(sb, "recovery complete");
327 nilfs_clear_recovery_info(&ri);
328 sb->s_flags = s_flags;
332 nilfs_err(sb, "error %d while searching super root", err);
336 iput(nilfs->ns_cpfile);
337 iput(nilfs->ns_sufile);
341 nilfs_clear_recovery_info(&ri);
342 sb->s_flags = s_flags;
346 static unsigned long long nilfs_max_size(unsigned int blkbits)
348 unsigned int max_bits;
349 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
351 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
353 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
358 * nilfs_nrsvsegs - calculate the number of reserved segments
359 * @nilfs: nilfs object
360 * @nsegs: total number of segments
362 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
364 return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
365 DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
369 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
371 nilfs->ns_nsegments = nsegs;
372 nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
375 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
376 struct nilfs_super_block *sbp)
378 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
379 nilfs_err(nilfs->ns_sb,
380 "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
381 le32_to_cpu(sbp->s_rev_level),
382 le16_to_cpu(sbp->s_minor_rev_level),
383 NILFS_CURRENT_REV, NILFS_MINOR_REV);
386 nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
387 if (nilfs->ns_sbsize > BLOCK_SIZE)
390 nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
391 if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
392 nilfs_err(nilfs->ns_sb, "too large inode size: %d bytes",
393 nilfs->ns_inode_size);
395 } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
396 nilfs_err(nilfs->ns_sb, "too small inode size: %d bytes",
397 nilfs->ns_inode_size);
401 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
403 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
404 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
405 nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
406 nilfs->ns_blocks_per_segment);
410 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
411 nilfs->ns_r_segments_percentage =
412 le32_to_cpu(sbp->s_r_segments_percentage);
413 if (nilfs->ns_r_segments_percentage < 1 ||
414 nilfs->ns_r_segments_percentage > 99) {
415 nilfs_err(nilfs->ns_sb,
416 "invalid reserved segments percentage: %lu",
417 nilfs->ns_r_segments_percentage);
421 nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
422 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
426 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
428 static unsigned char sum[4];
429 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
433 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
435 bytes = le16_to_cpu(sbp->s_bytes);
436 if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
438 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
440 crc = crc32_le(crc, sum, 4);
441 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
443 return crc == le32_to_cpu(sbp->s_sum);
446 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
448 return offset < ((le64_to_cpu(sbp->s_nsegments) *
449 le32_to_cpu(sbp->s_blocks_per_segment)) <<
450 (le32_to_cpu(sbp->s_log_block_size) + 10));
453 static void nilfs_release_super_block(struct the_nilfs *nilfs)
457 for (i = 0; i < 2; i++) {
458 if (nilfs->ns_sbp[i]) {
459 brelse(nilfs->ns_sbh[i]);
460 nilfs->ns_sbh[i] = NULL;
461 nilfs->ns_sbp[i] = NULL;
466 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
468 brelse(nilfs->ns_sbh[0]);
469 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
470 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
471 nilfs->ns_sbh[1] = NULL;
472 nilfs->ns_sbp[1] = NULL;
475 void nilfs_swap_super_block(struct the_nilfs *nilfs)
477 struct buffer_head *tsbh = nilfs->ns_sbh[0];
478 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
480 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
481 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
482 nilfs->ns_sbh[1] = tsbh;
483 nilfs->ns_sbp[1] = tsbp;
486 static int nilfs_load_super_block(struct the_nilfs *nilfs,
487 struct super_block *sb, int blocksize,
488 struct nilfs_super_block **sbpp)
490 struct nilfs_super_block **sbp = nilfs->ns_sbp;
491 struct buffer_head **sbh = nilfs->ns_sbh;
492 u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
493 int valid[2], swp = 0;
495 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
497 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
501 nilfs_err(sb, "unable to read superblock");
505 "unable to read primary superblock (blocksize = %d)",
507 } else if (!sbp[1]) {
509 "unable to read secondary superblock (blocksize = %d)",
514 * Compare two super blocks and set 1 in swp if the secondary
515 * super block is valid and newer. Otherwise, set 0 in swp.
517 valid[0] = nilfs_valid_sb(sbp[0]);
518 valid[1] = nilfs_valid_sb(sbp[1]);
519 swp = valid[1] && (!valid[0] ||
520 le64_to_cpu(sbp[1]->s_last_cno) >
521 le64_to_cpu(sbp[0]->s_last_cno));
523 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
531 nilfs_release_super_block(nilfs);
532 nilfs_err(sb, "couldn't find nilfs on the device");
538 "broken superblock, retrying with spare superblock (blocksize = %d)",
541 nilfs_swap_super_block(nilfs);
543 nilfs->ns_sbwcount = 0;
544 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
545 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
551 * init_nilfs - initialize a NILFS instance.
552 * @nilfs: the_nilfs structure
554 * @data: mount options
556 * init_nilfs() performs common initialization per block device (e.g.
557 * reading the super block, getting disk layout information, initializing
558 * shared fields in the_nilfs).
560 * Return Value: On success, 0 is returned. On error, a negative error
563 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
565 struct nilfs_super_block *sbp;
569 down_write(&nilfs->ns_sem);
571 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
573 nilfs_err(sb, "unable to set blocksize");
577 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
581 err = nilfs_store_magic_and_option(sb, sbp, data);
585 err = nilfs_check_feature_compatibility(sb, sbp);
589 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
590 if (blocksize < NILFS_MIN_BLOCK_SIZE ||
591 blocksize > NILFS_MAX_BLOCK_SIZE) {
593 "couldn't mount because of unsupported filesystem blocksize %d",
598 if (sb->s_blocksize != blocksize) {
599 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
601 if (blocksize < hw_blocksize) {
603 "blocksize %d too small for device (sector-size = %d)",
604 blocksize, hw_blocksize);
608 nilfs_release_super_block(nilfs);
609 sb_set_blocksize(sb, blocksize);
611 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
615 * Not to failed_sbh; sbh is released automatically
616 * when reloading fails.
619 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
620 nilfs->ns_blocksize = blocksize;
622 get_random_bytes(&nilfs->ns_next_generation,
623 sizeof(nilfs->ns_next_generation));
625 err = nilfs_store_disk_layout(nilfs, sbp);
629 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
631 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
633 err = nilfs_store_log_cursor(nilfs, sbp);
637 err = nilfs_sysfs_create_device_group(sb);
641 set_nilfs_init(nilfs);
644 up_write(&nilfs->ns_sem);
648 nilfs_release_super_block(nilfs);
652 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
655 sector_t seg_start, seg_end;
656 sector_t start = 0, nblocks = 0;
657 unsigned int sects_per_block;
661 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
662 bdev_logical_block_size(nilfs->ns_bdev);
663 for (sn = segnump; sn < segnump + nsegs; sn++) {
664 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
668 nblocks = seg_end - seg_start + 1;
669 } else if (start + nblocks == seg_start) {
670 nblocks += seg_end - seg_start + 1;
672 ret = blkdev_issue_discard(nilfs->ns_bdev,
673 start * sects_per_block,
674 nblocks * sects_per_block,
682 ret = blkdev_issue_discard(nilfs->ns_bdev,
683 start * sects_per_block,
684 nblocks * sects_per_block,
689 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
691 unsigned long ncleansegs;
693 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
694 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
695 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
696 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
700 int nilfs_near_disk_full(struct the_nilfs *nilfs)
702 unsigned long ncleansegs, nincsegs;
704 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
705 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
706 nilfs->ns_blocks_per_segment + 1;
708 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
711 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
714 struct nilfs_root *root;
716 spin_lock(&nilfs->ns_cptree_lock);
717 n = nilfs->ns_cptree.rb_node;
719 root = rb_entry(n, struct nilfs_root, rb_node);
721 if (cno < root->cno) {
723 } else if (cno > root->cno) {
726 refcount_inc(&root->count);
727 spin_unlock(&nilfs->ns_cptree_lock);
731 spin_unlock(&nilfs->ns_cptree_lock);
737 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
739 struct rb_node **p, *parent;
740 struct nilfs_root *root, *new;
743 root = nilfs_lookup_root(nilfs, cno);
747 new = kzalloc(sizeof(*root), GFP_KERNEL);
751 spin_lock(&nilfs->ns_cptree_lock);
753 p = &nilfs->ns_cptree.rb_node;
758 root = rb_entry(parent, struct nilfs_root, rb_node);
760 if (cno < root->cno) {
762 } else if (cno > root->cno) {
765 refcount_inc(&root->count);
766 spin_unlock(&nilfs->ns_cptree_lock);
775 refcount_set(&new->count, 1);
776 atomic64_set(&new->inodes_count, 0);
777 atomic64_set(&new->blocks_count, 0);
779 rb_link_node(&new->rb_node, parent, p);
780 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
782 spin_unlock(&nilfs->ns_cptree_lock);
784 err = nilfs_sysfs_create_snapshot_group(new);
793 void nilfs_put_root(struct nilfs_root *root)
795 struct the_nilfs *nilfs = root->nilfs;
797 if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
798 rb_erase(&root->rb_node, &nilfs->ns_cptree);
799 spin_unlock(&nilfs->ns_cptree_lock);
801 nilfs_sysfs_delete_snapshot_group(root);